A distributed storage system is realized by clustering a plurality of storage nodes each of which includes a plurality of storage devices and a server controlling writing and reading data to and from the plurality of storage devices. The distributed storage system has a feature that the scalability of the system according to an increase in data to be stored is high. As one type of such a distributed storage system, there has been an object storage system that handles data as an object.
Various kinds of techniques have been proposed that are used for improving performance of the storage system. For example, there has been a storage system that moves a logical volume within a storage subsystem to another storage subsystem so that loads are balanced between the storage subsystems. In addition, there has been a storage system where, in a fibre channel (FC) switch, connections between a plurality of ports individually connected to separate servers and a plurality of ports connected to redundant arrays of inexpensive disks (RAID) device are switched so that the loads of access paths are balanced.
Japanese Laid-open Patent Publication No. 2004-13215 and Japanese Laid-open Patent Publication No. 2007-304896 disclose related techniques.
Broadly speaking, examples of data to be stored in a distributed storage system include “hot” data frequently accessed, and “cold” data rarely accessed after having been stored once, such as data used for a backup. However, in many cases, at the time of storing data, it is not clear which data is hot data and which data is cold data, from among pieces of data to be stored. Therefore, if hot data is intensively stored in storage devices within the same storage node, the load of a server within the storage node becomes excessive, and hence, there has been a problem that performance for accessing individual storage devices connected to the server is significantly reduced.